Download pdf - Computer-Aided Volumetry of Pulmonary Nodules …qibawiki.rsna.org/images/0/0a/2009_Oda-ComputerAidedLung...ed at lung cancer screening with low radiation dose helical CT (low-dose

398 AJR:194, February 2010

high-resolution CT may help in differential diagnosis. Because the doubling time of BAC is long (average, 457–813 days) [8, 11], sub-jective evaluation of the growth rate by radi-ologists is unreliable.

With MDCT it is possible to scan a wide range, including areas containing pulmonary nodules, at a detector collimation of 0.500–0.625 mm in one breath-hold. This capability facilitates 3D evaluation of pulmonary nod-ules. In previous studies [12–16], investiga-tors assessed computer-aided volumetry of pulmonary nodules using volumetric data ob-tained at MDCT, and the technique report-edly had sufficiently high accuracy and re-producibility [12]. In those studies, however, only solid pulmonary nodules were evaluated with the software used, and computer-aided volumetry of GGO nodules reportedly was difficult [15]. We developed computer-aided volumetry software that can be used to mea-sure the volume not only of solid but also of

Computer-Aided Volumetry of Pulmonary Nodules Exhibiting Ground-Glass Opacity at MDCT

Seitaro Oda1 Kazuo Awai1 Kohei Murao2 Akio Ozawa2 Yumi Yanaga1 Koichi Kawanaka1 Yasuyuki Yamashita1

Oda S, Awai K, Murao K, et al.

1Department of Diagnostic Radiology, Graduate School of Medical Sciences, Kumamoto University, 1-1-1 Honjyo, Kumamoto 860-8556, Japan. Address correspondence to S. Oda.

2Bio-IT Business Development Group, Fujitsu, Tokyo, Japan.

Cardiopulmonar y Imaging • Or ig ina l Research

AJR 2010; 194:398–406

0361–803X/10/1942–398

© American Roentgen Ray Society

Among pulmonary nodules detect-ed at lung cancer screening with low radiation dose helical CT (low-dose CT), 19–38% exhibit

focal ground-glass opacity (GGO) [1, 2]. Nodules exhibiting GGO (GGO nodules) may be attributable to focal inflammation, fo-cal interstitial fibrosis [3], atypical adenoma-tous hyperplasia [4, 5], bronchioloalveolar carci noma (BAC) [4–6], or adenocarcinoma [4–6]. Although important, differentiation of BAC and adenocarcinoma, which are malig-nant, from other diseases can be difficult on a single CT scan [7]. Although many inflam-matory lesions resolve spontaneously or with antibiotic treatment [5], the size of GGO nodules attributable to BAC or adenocarci-noma gradually increases [8–10]. Focal in-terstitial fibrosis and atypical adenomatous hyperplasia with pure GGO remain stable in size for months or years [3]. Therefore, mon-itoring nodule size for several months with

Keywords: computer-aided diagnosis, ground-glass opacity, high-resolution CT, pulmonary nodules, volumetry

DOI:10.2214/AJR.09.2583

Received February 14, 2009; accepted after revision July 19, 2009.

The employment status of K. Murao and A. Ozawa at Fujitsu did not influence the data in this study.

OBJECTIVE. The purpose of this study was to investigate the accuracy and reproducibil-ity of results acquired with computer-aided volumetry software during MDCT of pulmonary nodules exhibiting ground-glass opacity.

MATERIALS AND METHODS. To evaluate the accuracy of computer-aided volum-etry software, we performed thin-section helical CT of a chest phantom that included simu-lated 3-, 5-, 8-, 10-, and 12-mm-diameter ground-glass opacity nodules with attenuation of –800, –630, and –450 HU. Three radiologists measured the volume of the nodules and calcu-lated the relative volume measurement error, which was defined as follows: (measured nodule volume minus assumed nodule volume ÷ assumed nodule volume) × 100. Two radiologists performed two independent measurements of 59 nodules in humans. Intraobserver and inter-observer agreement was evaluated with Bland-Altman methods.

RESULTS. The relative volume measurement error for simulated ground-glass opacity nodules measuring 3 mm ranged from 51.1% to 85.2% and for nodules measuring 5 mm or more in diameter ranged from –4.1% to 7.1%. In the clinical study, for intraobserver agree-ment, the 95% limits of agreement were –14.9% and –13.7% and –16.6% to 15.7% for observ-ers A and B. For interobserver agreement, these values were –16.3% to 23.7% for nodules 8 mm in diameter or larger.

CONCLUSION. With computer-aided volumetry of ground-glass opacity nodules, the relative volume measurement error was small for nodules 5 mm in diameter or larger. In-traobserver and interobserver agreement was relatively high for nodules 8 mm in diameter or larger.

Oda et al.MDCT and Volumetry of Pulmonary Nodules

Cardiopulmonary ImagingOriginal Research

AJR:194, February 2010 399

MDCT and Volumetry of Pulmonary Nodules

GGO nodules. The purpose of this study was to investigate the accuracy and reproducibil-ity of results obtained on GGO nodules with our computer-aided volumetry software.

Materials and MethodsComputerized Volumetry of Pulmonary Nodules

For volumetry of GGO nodules (Figs. 1 and 2), we used prototype software developed for a PACS (3D analysis software for HOPE/DrABLE-EX, Fu-jitsu). First, with a single click-and-drag maneuver, the radiologists roughly specified the region of in-terest to include the target nodule on axial, sagittal, and coronal images. In an improvement of our pre-viously reported method [17], the nodule was auto-matically segmented with the computer-aided vol-umetry software. For robust extraction of nodules from the lung parenchyma, we adopted an adaptive threshold attenuation value based on the attenua-tion value of the target nodule and the lung paren-chymal background. When the difference between the attenuation value of the target nodule and that of the lung parenchyma was relatively large, the com-puter-aided volumetry software chose the threshold attenuation value that approximated the mean value of the nodule and lung parenchyma. When the dif-ference in the attenuation value of the nodule and lung parenchyma was relatively small, the thresh-old attenuation value nonlinearly approached the average attenuation value of the lung parenchyma.

In the pulmonary nodule extraction process, structures connected to the nodule, such as ves-sels and bronchi, were roughly eliminated by the

computer-aided volumetry software. Occasional-ly, some structures remained around the nodules (Fig. 2C), or part of the nodule was not included in the extracted area. In such cases, the radiologists who performed volumetry used the mathematic morphology erosion–dilation technique to edit the segmented area to include the nodule by chang-ing the threshold level that separated the density of the nodule from the base density of the lung [18] (Fig. 2D). Erosion–dilation is a useful digital image-processing technique for eliminating struc-tures tangent with the nodule, such as vessels and thoracic wall. Because it was based on the subjec-tive judgment of each radiologist, this modifica-tion resulted in volumetric intraobserver and in-terobserver discrepancies.

Although the prototype software allows free-hand shape editing, the radiologists were not al-lowed to use this technique. Rather, they could use only the aforedescribed image-processing technique to assure the reproducibility of segmen-tation. The judgment of successful segmentation was based on the observers’ visual assessment on axial CT images and on sagittal and coronal mul-tiplanar reconstructed images. Finally, the volume of the segmented area was automatically mea-sured with the software. The average calculation time required by the computer-aided volumetry system to analyze each nodule was less than 1.0 seconds. A computer workstation (FMV-H8230, Fujitsu) with dual 2.16-GHz processors (Core Duo, Intel) was used in this study.

Specification of ROI by radiologists

Density analysis of nodule and lung parenchyma

Automatic estimation of nodule border

Subjective modification of nodule border

Calculation of nodule volume

Fig. 1—Diagram of computerized scheme for volumetry of nodules on thin-section helical CT images. ROI = region of interest.

A

Fig. 2—62-year-old woman with bronchioalveolar carcinoma. Computerized scheme for volumetry of nodules on thin-section helical CT images.A, High-resolution CT scan shows target ground-glass opacity nodule.B, High-resolution CT scan shows target ground-glass opacity nodule specified by and place region of interest manually placed by radiologists.C, High-resolution CT scan shows density of nodule and surrounding lung parenchyma automatically analyzed with software and nodule border estimated. Structures such as vessels occasionally remained around nodules (arrows).D, High-resolution CT scan shows nodule border subjectively modified by radiologists using semiautomatic edit tool and concept of mathematic morphology. Nodule volume is automatically calculated with software.

B

C D


Oda et al.

Phantom StudyChest phantom—To evaluate the accuracy of

our computer-aided volumetry software, we con-ducted a phantom study with simulated GGO nod-ules. Ours was a commercially available chest CT phantom with simulated GGO nodules (multipur-pose chest phantom N1, Kyoto Kagaku). In this phantom, simulated soft tissues, such as pulmo-nary vessels, the chest wall, heart, diaphragm, and liver, consist of polyurethane resin composites, and simulated bone consists of an epoxide resin. The space between the pulmonary vessels, heart, and chest wall was filled with air. The chest wall can be removed from the other structures, such as the simulated heart, pulmonary vessels, diaphragm, and liver. Simulated nodules were spheres made of urethane foam resin. The diameters of the sim-ulated nodules were 3 mm (estimated error, 20%), 5 mm (estimated error, 6%), 8 mm (estimated er-ror, 4%), 10 mm (estimated error, 3%), and 12 mm (estimated error, 2.5%), and the attenuation val-ues of the simulated nodules were –800, –630, and –450 HU. On CT scans of the chest phantom har-boring simulated nodules, we placed a nodule on bifurcations of the pulmonary vessels in the right lung at the level of the carina.

CT of phantom—CT was performed with a 64-MDCT scanner (Brilliance 64, Philips Health-care). The scanning parameters were detector row width, 0.625 mm; helical pitch, 0.673; sec-tion thickness and section interval, 1 mm; rotation time, 0.5 second; tube voltage, 120 kVp; tube cur-rent, 250 mAs.

Volumetry of simulated GGO nodules—Figure 3 shows the screen display of the computer-aided volumetry software in the phantom study. Three radiologists with 4-, 10-, and 15 years of experi-ence in chest CT subjected each simulated nodule to three volumetric measurements, and the resul-tant values were averaged for each nodule. Each radiologist performed three volumetry sessions at 1-week or longer intervals. We calculated the rela-tive volume measurement error for each nodule to evaluate the accuracy of our volumetry software. The error was defined as follows: [(measured nod-ule volume – assumed nodule volume) ÷ assumed nodule volume] × 100. The assumed nodule vol-ume was the calculated volume based on the di-ameter of a simulated spherical nodule. We calcu-lated the mean relative volume measurement error for each nodule by averaging the values reported by the three radiologists.

Clinical StudyTo evaluate the reproducibility of our computer-

aided volumetry software, represented by intraob-server and interobserver agreement, we conducted a study using data derived from patients. We also

investigated whether the need for editing the seg-ment area or the edit time depends on the mor-phologic features of the GGO nodule. All patients who underwent CT examinations at our institu-tion had given prior informed consent for the use of their CT images in future retrospective studies. Our institutional review board approved the use of the CT database. The requirement for informed specific study-related consent was waived.

Nodule selection—One chest radiologist with 21 years of chest CT experience reviewed the re-cords of 211 consecutively registered patients with suspected pulmonary nodules. The patients un-derwent thin-section helical CT of the chest at our institution during the 36-month period from Janu-ary 2004 through December 2006. The radiolo-gist, who did not participate in the volume mea-surement study selected all patients who satisfied the following criteria: GGO nodules that did not exceed 20 mm in the xy (transverse) plane, a his-tologic diagnosis based on findings at thoracic surgery or at CT-guided transcutaneous or bron-choscopic transbronchial biopsy, and absence of consolidation due to the presence of organizing tissues after pneumonia or associated with idio-pathic pulmonary fibrosis around the nodules. The last criterion was applied because we found it dif-ficult to define the nodule boundary.

On the basis of the selection criteria, 59 nod-ules in 59 patients (18 men, 41 women; mean age, 65.4 years; range, 47–79 years ) with 40 malignant (BAC, n = 38; adenocarcinoma, n = 2) and 19 be-nign nodules (atypical adenomatous hyperplasia, n = 17; amyloidosis, n = 1; goblet cell metapla-sia, n = 1) were identified. The mean size in the xy (transverse) plane of the 59 nodules was 12.8 ± 4.7 (SD) mm (range, 4.0–20.0 mm). The mean at-tenuation was –539.6 ± 96.3 HU (range, –744 to –339 HU).

CT—CT scans were obtained with a 4-MDCT scanner (LightSpeed QX/I, GE Healthcare). After routine helical scanning of the whole thorax, thin-slice helical scans with a scan range of approximate-ly 5 cm including the pulmonary nodules were ob-tained. The scan parameters for routine helical scans were detector collimation, 4 × 2.5 mm; helical pitch (beam pitch), 1.5; slice thickness and interval, 5.0 mm; rotation time, 0.8 second; 120 kVp; 250 mA. The scan parameters for thin-slice helical scans were detector row width, 4 × 1.25 mm; helical pitch, 0.75; slice thickness and interval, 1.25 mm; rotation time, 0.8 seconds; 120 kVp; 160–200 mA. The recon-struction algorithm for thin-slice helical scans was bone plus. Contrast enhancement was not used in any of the 59 nodules. Because the 4-MDCT scanner was replaced with the 64-MDCT scanner in January 2008, we used the 4-MDCT scanner for the clinical and the 64-MDCT scanner for the phantom study.

Volumetry of GGO nodules—Figure 4 shows the screen display of computer-aided volumetry soft-ware in the clinical study. Two of the three radiolo-gists (4 and 15 years of experience with chest CT) who participated in the phantom study performed volumetry in the clinical study. Each radiologist performed two volumetry sessions at 1-month or longer intervals. Between the two sessions, the or-der of case presentation was changed to minimize the learning effect. Before the first session, each ra-diologist underwent a training session that involved volumetry in three training cases to become famil-iar with the editing tools of the computer-aided volumetry software because the shape of some of the human nodules was more complex than that of the phantom nodules. The three training cases were not among the 59 cases in the clinical study.

When the GGO nodules were not accurately ex-tracted with the computer-aided volumetry software alone, the two radiologists subjectively edited the

Fig. 3—Chest phantom. Screen display of computer-aided volumetry software used in phantom study shows axial, coronal, and sagittal images and maximum intensity projections. Simulated ground-glass opacity nodule (–630 HU) can be accurately extracted with software.



segmented area using techniques based on mathe-matic morphology. If an edit was required, we mea-sured the editing time. To investigate whether mor-phologic nodule characteristics affected the need for editing, the chest radiologist who selected the 59 pa-tients classified the GGO nodules into three catego-ries on the basis of the nature of the nodule margin. Category 1 nodules (n = 22) exhibited a well-defined smooth margin; category 2 nodules (n = 24) had a well-defined irregular margin; and category 3 nod-ules (n = 13) manifested an ill-defined margin. The radiologist also classified the 59 nodules as partly solid (mixed GGO, n = 31) and nonsolid (pure GGO, n = 28) according to internal density [1]. The readers recorded whether the GGO nodules were (n = 12) or were not (n = 47) adjacent to the pleural surface.

Statistical AnalysisAll numerical values are reported as mean ± SD.

We used Bland-Altman [19] analysis to determine intraobserver and interobserver agreement in the clinical study. We assessed intraobserver and inter-observer agreement on nodules < 8 mm and ≥ 8 mm in diameter to evaluate the influence of nodule size. We also assessed agreement on nodules that were or were not subjected to editing to reveal wheth-er and how editing affected the reproducibility of computer-aided volumetry. On the Bland-Altman plots we used the percentage difference of the av-erage of two measured values (relative difference) rather than absolute difference because the vari-ability of the differences increased as the magni-tude of the measurements increased.

To determine the frequency of necessary edits (edit rate) required by the morphologic features of the GGO nodules, we performed the chi-square test. To assess the effect of the morphologic fea-tures of nodules on edit time, we used Student’s t test or Tukey multiple comparisons. Variables with a value of p < 0.05 were considered to in-dicate statistically significant differences. Statis-tical analysis software was used (MedCalc Soft-ware, MedCalc; SPSS version 15.0, SPSS).

ResultsPhantom Study

The average of the relative volume mea-surement error values for nodules with a di-ameter of 3 mm and attenuation of –800, –630, and –450 HU was 61.5%, 51.1%, and 85.2% (Fig. 5). For all nodules 5 mm in di-ameter or larger, the average ranged from –4.1% to 7.1%. The SD of relative volume measurement error for 3-mm diameter nod-ules with attenuation values of –800, –630, and –450 HU was 4.0%, 9.8%, and 13.5%. It was 0.9% to 6.2% for all nodules with a di-ameter of 5 mm or more.

Clinical StudyWe were able to perform computer-aid-

ed volumetry on all 59 GGO nodules in pa-tients. The mean intraobserver agreement on all GGO nodules was 1.6 ± 11.1% (95% CI, –1.3% to 4.5%) and 1.8 ± 11.5% (95% CI, –1.2% to 4.8%) for observers A and B

(Fig. 6, Tables 1 and 2). The mean interob-server agreement for all GGO nodules was 3.2 ± 11.8% (95% CI, 0.1–6.3%) (Fig. 7, Ta-bles 1 and 2). Observer A applied editing to 36 nodules (61.0%) and observer B to 28 nod-ules (47.5%). Table 3 shows the edit rates for three categories of GGO nodules for the two observers. There was a statistically significant difference in the edit rate of the three catego-ries. There was no statistically significant dif-ference in the edit rates of nonsolid and partly solid GGO nodules. The edit rate for nodules that were adjacent to the pleura was signifi-cantly higher than the rate for nodules not ad-jacent to the pleura.

The mean edit times were 24.1 ± 10.0 sec-onds and 35.3 ± 12.3 seconds for observers A and B (Table 4). The mean edit time for cat-egory 3 nodules was statistically significant-ly longer than for categories 1 and 2 nodules (p < 0.01 and < 0.01 for observer A; p < 0.01 and < 0.01 for observer B). The difference in the edit times of the observers was not sta-tistically significant for categories 1 and 2 nodules (p = 0.92 and p = 0.61). For observ-er B the time to edit partly solid GGOs was statistically significantly longer than that to edit nonsolid GGOs (p < 0.01); for observer A there was no statistically significant differ-ence (p = 0.36). The times to edit nodules ad-jacent to and not adjacent to pleura were not significantly different (observer A, p = 0.77; observer B, p = 0.35).

−202 4

Rel

ativ

e Vo

lum

e M

easu

rem

ent

Err

or

(%)

0

60

80

100−450 HU

−800 HU−630 HU

40

20

6 8Nodule Diameter (mm)

10 12 14

Fig. 4—57-year-old woman with atypical adenomatous hyperplasia. Screen display shows accurate extraction of pure ground-glass opacity nodule (category 1) with computer-aided volumetry software.

Fig. 5—Graph shows average relative volume measurement error for nodules measuring 3, 5, 8, 10, and 12 mm in diameter with attenuation of –450, –630, and –800 HU. For nodules with attenuation of –450 HU, relative volume measurement error was 85.2% ± 13.5% (SD), –4.1% ± 3.4%, 4.8% ± 2.6%, 1.7% ± 3.4%, and –1.5% ± 4.2%. For nodules with attenuation of –630 HU, relative volume measurement error was 51.1% ± 9.8%, 0.2% ± 6.2%, 3.3% ± 1.4%, 0.2% ± 3.3%, and 1.6% ± 3.7%. For nodules with attenuation of –800 HU, relative volume measurement error was 61.5% ± 4.0%, 7.1% ± 2.1%, 4.8% ± 1.4%, 3.1% ± 1.2%, and 7.0% ± 0.9%.


Oda et al.

DiscussionRegardless of difficulty in the volumetry

of GGO nodules, all GGO nodules in this study were successfully segmented with our computer-aided volumetry software. Our re-sults were better than previously reported volumetry measurements of solid nodules in humans [20, 21]; in those studies, 71–97% of nodules were successfully segmented.

Factors that affect computer-aided volu-metry of pulmonary nodules are the algo-rithm of the volumetry software, the thresh-old values for nodule extraction, nodule size and attenuation, tube current, and the image reconstruction kernel [15, 16, 22, 23]. The

presence of pulmonary vessels, bronchi, or chest wall connecting with the nodule may require manual modification of nodule ex-traction, and arbitrary manipulation by radi-ologists may be a factor affecting the accura-cy of volumetry. De Hoop et al. [20] assessed the volumetry of solid nodules with six semi-automated software packages. They found manual modification of nodule extraction improved the accuracy of volumetry without significantly affecting reproducibility.

In the evaluation of solid pulmonary nod-ules, computer-aided volumetry produced only minimal errors and had high measure-ment reproducibility [12, 13, 23, 24]. Kos-

tis et al. [23], who evaluated computer-aid-ed volumetry of solid nodules in humans, reported that the overall standard measure-ment error for nodules 2–5 mm, 5–8 mm, and 8–10 mm was 18.5%, 10.6%, and 7.47%. A study conducted by Das et al. [12] revealed that accurate volumetry was possible for sim-ulated solid nodules with a diameter of 5 mm or larger and with a volume greater than 66 mm3. However, volumetry of GGO nodules is difficult because the attenuation difference between these nodules and normal lung pa-renchyma is small, and other structures in the lung are visualized in the same attenu-ation range. An earlier volumetric study

−50

50

Rel

ativ

e D

iffe

ren

ce (

%)

−40

−40 −25.3−1.96 SD

7.4Mean

40.11.96 SD

−20

−10

0

10

20

30

40

50

0 100 150

Mean Volume (mm3)

200 250 300

−50

500

Rel

ativ

e D

iffe

ren

ce (

%)

−40

−40

−14.9−1.96 SD

−0.6Mean

13.71.96 SD

−20

−10

0

10

20

30

40

50

0 1,000 1,500

Mean Volume (mm3)

2,5002,000 3,000 3,500

A

Fig. 6—Intraobserver agreement.A, Scatterplot shows mean intraobserver agreement on nodules < 8 mm in diameter for observer A was 7.4% ± 16.7% (95% CI, –1.5% to 16.3%); 95% limits of agreement were –25.3% (95% CI, –40.8 to –9.8%) and 40.1% (95% CI, 24.6–55.6%).B, Scatterplot shows mean intraobserver agreement on nodules ≥ 8 mm in diameter for observer A was –0.6% ± 7.3% (95% CI, –2.8% to 1.7%); 95% limits of agreement were –14.9% (95% CI, –18.7% to –11.0%) and 13.7% (95% CI, 9.8–17.6%).C, Scatterplot shows mean intraobserver agreement on nodules < 8 mm in diameter for observer B was 7.0% ± 15.8% (95% CI, –0.9% to 14.9%); 95% limits of agreement were –24.0% (95% CI, –37.7% to –10.3%) and 38.0% (95% CI, 24.3–51.7%).D, Scatterplot shows mean intraobserver agreement on nodules ≥ 8 mm in diameter for observer B was –0.4% ± 8.2% (95% CI, –3.0% to 2.2%); 95% limits of agreement were –16.6% (95% CI, –21.1% and –12.1%) and 15.7% (95% CI, 11.2%–20.2%).

B

−50

50

Rel

ativ

e D

iffe

ren

ce (

%)

−40

−40 −24.0−1.96 SD

7.0Mean

38.01.96 SD

−20

−10

0

10

20

30

40

50

0 100 150

Mean Volume (mm3)

200 250 300

−50

500

Rel

ativ

e D

iffe

ren

ce (

%)

−40

−40

−16.6−1.96 SD

−4.0Mean

15.71.96 SD

−20

−10

0

10

20

30

40

50

1,0000 1,500 2,000

Mean Volume (mm3)

2,500 3,000 3,500

C D



[15] on chest phantoms showed significantly higher volume measurement error for GGO than for solid nodules.

In our phantom study, we obtained accu-rate volumetric measurements for nodules larger than 5 mm in diameter; the mean mea-surement error for nodules 5 mm or larger was 2.35% (range, –4.14% to 7.13%). Our ac-curacy rate was equivalent to that in previ-ous studies of solid nodules. In the clinical setting, however, pulmonary nodules are not necessarily round, their margins are not al-ways clear, and they are frequently in contact with adjacent structures, such as the chest wall, pulmonary vessels, and bronchi. In ad-dition, artifacts due to respiratory and cardiac motion are often present on clinical images. The simulated nodules in the chest phantom, however, were round and had clear margins, and there were no motion artifacts. Volumet-ric measurement therefore may be less accu-rate in clinical than in phantom studies. We observed measurement variability even in volumetry of simulated GGO nodules in this

study. Because the software defines the opti-mal threshold attenuation within each region of interest specified by the radiologist, the threshold value varies with the size or site of the region of interest. In addition, in some instances editing was needed even for simu-lated GGO nodules. This factor may account for the measurement variability encountered in our phantom study.

In our clinical investigation, intraobserv-er and interobserver agreement was relative-ly high for nodules ≥ 8 mm in diameter. For intraobserver agreement, the 95% limits of agreement were –14.9% and 13.7% for ob-server A and –16.6% and 15.7% for observer B. For interobserver agreement, the limits of agreement were –16.3% and 23.7% for nod-ules ≥ 8 mm in diameter. Wormanns et al. [13] obtained excellent results in a study in which intraobserver agreement was –3.9% to 5.7% and interobserver agreement was –5.5% to 6.6% for automated volumetry of clinical solid nodules. Thus volumetric measure-ments of GGO nodules exhibited larger vari-

ability than did solid nodules. Taking into account our intraobserver and interobserver agreement results, the threshold for identify-ing an increase in the measured volume of a GGO nodule is a 30% increase, equivalent to a 9% increase in nodule diameter. For ex-ample, if a 10-mm diameter of a nodule on a baseline scan has increased 1 mm on a fol-low-up scan, it can be difficult to detect this change with visual assessment. Therefore, we conclude that our results on intraobserv-er and interobserver agreement are clinically acceptable for early detection of growth of GGO nodules.

We found that intraobserver and interob-server agreement was slightly better for nod-ules without than for those with editing. We posit, however, that this difference is minimal and acceptable for clinical practice. Although editing based on the subjective judgment of each observer resulted in a few intraobserver and interobserver discrepancies, appropriate editing may be indispensable for accurate volumetry of GGO nodules.

TABLE 1: Comparison of Intraobserver and Interobserver Agreement

Nodules Evaluated Mean Upper Limit Lower Limit

All nodules

Intraobserver agreement observer A 1.6 (–1.3 to 4.5) 23.3 (18.4 to 28.3) –20.2 (–25.2 to –15.2)

Intraobserver agreement observer B 1.8 (–1.2 to 4.8) 24.4 (19.2 to 29.5) –20.7 (–25.8 to –15.5)

Interobserver agreement 3.2 (0.1 to 6.3) 26.4 (21.1 to 31.7) –20.0 (–25.3 to –14.7)

Nodules < 8 mm in diameter

Intraobserver agreement observer A 7.4 (–1.5 to 16.3) 40.1 (24.6 to 55.6) –25.3 (–40.8. to –9.8)

Intraobserver agreement observer B 7.0 (–0.9 to 14.9) 38.0 (24.3 to 51.7) –24.0 (–37.7 to –10.3)

Interobserver agreement 1.9 (–6.0 to 9.8) 32.2 (18.3 to 46.0) –28.4 (–42.2 to –14.5)

Nodules ≥ 8 mm in diameter

Intraobserver agreement observer A –0.6 (–2.8 to 1.7) 13.7 (9.8 to 17.6) –14.9 (–18.7 to –11.0)

Intraobserver agreement observer B –0.4 (–3.0 to 2.2) 15.7 (11.2 to 20.2) –16.6 (–21.1 to –12.1)

Interobserver agreement 3.7 (0.5 to 6.9) 23.7 (18.3 to 29.2) –16.3 (–21.8 to –10.8)

Note—Values are percentages. Values in parentheses are 95% CI.

TABLE 2: Influence of Editing on Intraobserver and Interobserver Agreement

Use of Editing Mean Upper Limit Lower Limit

Nodules without editing

Intraobserver agreement observer A 2.3 (−1.2 to 5.9) 18.3 (12.2 to 24.4) −13.6 (−19.7 to −7.5)

Intraobserver agreement observer B 0.1 (−3.1 to 3.3) 17.3 (11.7 to 22.8) −17.1 (−22.6 to −11.5)

Interobserver agreement 3.5 (−0.3 to 7.2) 20.1 (13.6 to 26.6) −13.1 (−19.6 to −6.6)

Nodules with editing

Intraobserver agreement observer A 1.1 (−3.2 to 5.4) 26.0 (18.6 to 33.5) −23.9 (−31.3 to −16.4)

Intraobserver agreement observer B 3.8 (−1.6 to 9.1) 30.8 (21.6 to 40.1) −23.3 (−32.6 to −14.1)

Interobserver agreement 2.7 (−2.8 to 8.2) 28.2 (18.7 to 37.8) −22.9 (−32.4 to −13.4)

Note—Values are percentages. Values in parentheses are 95% CI.


Oda et al.

In our clinical study, edits were needed for approximately one half of GGO nodules. For category 3 GGO nodules, the edit rate was significantly higher and the edit time sig-nificantly longer than for categories 1 and 2 nodules. However, for both observers, the mean edit time for category 3 GGO nodules was shorter than 60 seconds. Furthermore, there was no statistically significant differ-ence in edit time between nodules that were and those that were not adjacent to the pleu-ra. The mean edit time was approximately 30 seconds for both pleura-adjacent and nonad-jacent GGO nodules, although the edit rate for nodules adjacent to the pleura was high-

er. Considering the acceptable length of edit time, we suggest that our computer-aided volumetry software is a practical tool for the measurement of GGO nodules in humans.

One of the most important indicators of malignancy is the growth rate of nodules, commonly expressed as the volume doubling time [13]. Computer-aided volumetric as-sessment of pulmonary nodules to estimate their growth rate has gained wider accep-tance [14, 16]. The reported average BAC doubling time, calculated from the maximal tumor diameter with the Schwartz equation [25] is 457–813 days [8, 11]. In the interpre-tation of these data, it must be remembered

that the reported doubling times for GGO nodules were not based on 3D data. Accord-ing to Marten et al. [24], the relative mean error of volumetric measurements of pul-monary nodules is significantly higher for manual unidimensional measurements. In-traobserver and interobserver agreement on manual unidimensional measurements was lower than with automated volumetry. Yan-kelevitz et al. [16] also found that comput-er-aided 3D volumetry had clear advantages over conventional bilinear 2D measurement for accurate representation of nodular vol-ume. We recommend that the doubling time of pulmonary nodules be reevaluated with highly accurate computer-aided volumetry software that features high reproducibility.

There are inherent problems with volum-etry of GGO nodules. First, the size of some BACs with GGO decreased in the course of follow-up owing to the collapse of alveolar spaces, fibrosis, or severe narrowing of the al-veolar space [10]. Therefore, even if the size of GGO nodules decreases, malignancy can-not be ruled out. Second, because the cells of BAC with focal GGO manifest replacement growth without marked reduction in the al-veolar lumen and without marked thicken-ing of alveolar septa [26], BAC with focal GGO is a relatively soft tumor. Therefore, the volume of BAC with focal GGO may be affected by the lung volume. Furthermore, computer-aided volumetry of GGO nodules seems to be affected by a change in the atten-uation of the lung surrounding parenchyma that occurs with inspiratory level. In volume-

Fig. 7—Interobserver agreement.A, Scatterplot shows mean interobserver agreement on nodules < 8 mm in diameter was 1.9% ± 15.4% (95% CI, –6.0 to 9.8%); 95% limits of agreement were –28.4% (95% CI, –42.2% to –14.5%) and 32.2% (95% CI, 18.3%–46.0%).B, Scatterplot shows mean interobserver agreement on nodules ≥ 8 mm in diameter was 3.7% ± 10.2% (95% CI, 0.5%–6.9%); 95% limits of agreement were –16.3% (95% CI, –21.8% to –10.8%) and 23.7% (95% CI, 18.3%–29.2%).

−50

50

Rel

ativ

e D

iffe

ren

ce (

%)

−40

−40 −28.4−1.96 SD

1.9Mean

32.21.96 SD

−20

−10

0

10

20

30

40

50

0 100 150

Mean Volume (mm3)

200 250 300

−50

500

Rel

ativ

e D

iffe

ren

ce (

%)

−40

−40

−16.3−1.96 SD

3.7Mean

23.71.96 SD

−20

−10

0

10

20

30

40

50

0 1,000 1,500

Mean Volume (mm3)

2,5002,000 3,000 3,500

A B

TABLE 3: Edit Rate for Ground-Glass Opacity Nodules According to Morphologic Subtype

Ground-Glass Opacity Subtype

Observer A Observer B

Edit Rate (%) pa Edit Rate (%) p

Margin characteristics 0.03 0.04

Category 1 (22) 54.5 (12) 31.8 (7)

Category 2 (24) 50.0 (12) 45.8 (11)

Category 3 (13) 92.3 (12) 76.9 (10)

Internal density 0.40 0.53

Nonsolid (28) 53.6 (15) 53.6 (15)

Partly solid (31) 67.7 (21) 41.9 (13)

Relation to pleura 0.04 0.03

Adjacent to pleura (12) 91.7 (11) 83.3 (9)

Not adjacent to pleura (47) 53.2 (25) 38.3 (18)

Total (n = 59) 61.0 (36) 47.5 (28)

Note—Numbers in parentheses are number of patients.aChi-square test.



try of focal GGO, the lung volume may have to be taken into consideration.

Debate continues about the treatment of pa-tients with GGO nodules. To our knowledge, no consensus has been reached on whether GGO nodules should be resected or observed. At present there is no definite evidence on the natural history of GGO nodules.

Evaluating the volume-doubling time of GGO nodules with accurate computer-aided volumetry may yield information on tumor activity and lead to better management of GGO nodules. Routine computer-aided volu-metry in the follow-up of GGO nodules may also be beneficial. Ikeda et al. [27], who per-formed 3D analysis of the attenuation within GGO nodules using computer-aided volum-etry software for differentiating atypical ad-enomatous hyperplasia, BAC, and adenocar-cinoma, found this method useful. However, even these 3D analyses require high-preci-sion computer-aided volumetry.

During the follow-up of GGO nodules, sol-id components may appear or grow. Accord-ing to Henschke et al. [1], the malignancy rate for partly solid GGO nodules was 63%. In ad-enocarcinoma with partly solid nodules, the ratio of solid to nonsolid parts is related to the prognosis. It is also useful for differentiation of adenocarcinoma subtypes [6, 9]. Therefore, a function for the identification and quantifi-cation of solid components within GGO nod-ules should be added to computer-aided diag-nosis at CT follow-up of GGO nodules.

There were several potential limitations to our study. First, because we used a 64-MDCT scanner in the phantom study and a 4-MDCT scanner in the clinical study, the

results of the phantom study may not be ap-plicable to the clinical study. Das et al. [12, 28], who compared the accuracy of automat-ed volumetry of solid phantom nodules us-ing CT scanners from different vendors and scanners with different numbers of detectors, concluded that solid nodule volumetry was accurate and that the degree of volume error was acceptable considering that the data were acquired using different scanners. For GGO nodules, however, volumetry with different scanners may lead to a change in the cutoff value and turn out to be a source of variabil-ity. A technique that unifies and rectifies the image background between types of scan-ners is needed. Second, we used a section thickness and interval of 1 mm in the phan-tom study. In the clinical study, these values were 1.25 mm because the clinical data were collected in a retrospective manner. Goo et al. [29] observed a tendency toward larger volume measurement errors with increasing section thickness. Thus the measured volume in our clinical study may have been overesti-mated. Third, we did not evaluate interscan variability. Wormanns et al. [13], who used computer-aided volumetry to assess solid pulmonary nodules on two consecutive CT scans on the same day, reported that the 95% limits of agreement were –20.4% and 21.9% (standard error, 1.5%). Gietema et al. [30] also reported similar limits of agreement for interscan variability in volumetry of solid nodules. They suggested that most of the ob-served measurement variability was attrib-utable to interscan differences. This finding indicates that the effect of interscan variabil-ity must be assessed. Last, there may have

been selection bias in this study because we selected patients on the basis of the strict cri-teria described earlier.

We conclude that with our computer-aid-ed volumetry software, volumetry was rel-atively accurate and that intraobserver and interobserver agreement may be clinically acceptable for the early detection of growth in GGO nodules ≥ 8 mm in diameter. On the basis of our findings we suggest that accu-rate computer-aided volumetry can play an important role in the follow-up and manage-ment of GGO nodules.

References 1. Henschke CI, Yankelevitz DF, Mirtcheva R,

McGuinness G, McCauley D, Miettinen OS. CT

screening for lung cancer: frequency and signifi-

cance of part-solid and nonsolid nodules. AJR

2002; 178:1053–1057

2. Li F, Sone S, Abe H, MacMahon H, Armato SG

3rd, Doi K. Lung cancers missed at low-dose heli-

cal CT screening in a general population: com-

parison of clinical, histopathologic, and imaging

findings. Radiology 2002; 225:673–683

3. Park CM, Goo JM, Lee HJ, et al. Focal interstitial

fibrosis manifesting as nodular ground-glass

opacity: thin-section CT findings. Eur Radiol

2007; 17:2325–2331

4. Nakajima R, Yokose T, Kakinuma R, Nagai K,

Nishiwaki Y, Ochiai A. Localized pure ground-

glass opacity on high-resolution CT: histologic

characteristics. J Comput Assist Tomogr 2002;

26:323–329

5. Nakata M, Saeki H, Takata I, et al. Focal ground-

glass opacity detected by low-dose helical CT.

Chest 2002; 121:1464–1467

6. Kuriyama K, Seto M, Kasugai T, et al. Ground-

glass opacity on thin-section CT: value in differ-

entiating subtypes of adenocarcinoma of the lung.

AJR 1999; 173:465–469

7. Kim HY, Shim YM, Lee KS, Han J, Yi CA, Kim

YK. Persistent pulmonary nodular ground-glass

opacity at thin-section CT: histopathologic com-

parisons. Radiology 2007; 245:267–275

8. Aoki T, Nakata H, Watanabe H, et al. Evolution of

peripheral lung adenocarcinomas: CT findings

correlated with histology and tumor doubling

time. AJR 2000; 174:763–768

9. Aoki T, Tomoda Y, Watanabe H, et al. Peripheral

lung adenocarcinoma: correlation of thin-section

CT findings with histologic prognostic factors and

survival. Radiology 2001; 220:803–809

10. Kakinuma R, Ohmatsu H, Kaneko M, et al. Pro-

gression of focal pure ground-glass opacity de-

tected by low-dose helical computed tomography

screening for lung cancer. J Comput Assist Tomogr

2004; 28:17–23

TABLE 4: Edit Time for Ground-Glass Opacity Nodules According to Morphologic Subtype

Ground-Glass Opacity Subtype

Observer A Observer B

Edit Time (s) pEdit Time

(s) p

Margin characteristics

Category 1 (22) 17.5 ± 5.2 Category 1 vs 2, 0.92 29.6 ± 9.9 Category 1 vs 2, 0.61



Internal density 0.36 0.01

Nonsolid (28) 22.2 ± 8.8 29.5 ± 8.8

Partly solid (31) 25.4 ± 10.9 41.9 ± 12.7

Relation to pleura 0.77 0.35

Adjacent to pleura (12) 24.8 ± 8.0 32.3 ± 9.9

Not adjacent to pleura (47) 23.7 ± 11.0 36.3 ± 13.4

Total (n = 59) 24.1 ± 10.0 35.3 ± 12.3


Oda et al.

11. Hasegawa M, Sone S, Takashima S, et al. Growth

rate of small lung cancers detected on mass CT

screening. Br J Radiol 2000; 73:1252–1259

12. Das M, Ley-Zaporozhan J, Gietema HA, et al. Ac-

curacy of automated volumetry of pulmonary

nodules across different multislice CT scanners.

Eur Radiol 2007; 17:1979–1984

13. Wormanns D, Kohl G, Klotz E, et al. Volumetric

measurements of pulmonary nodules at multi-row

detector CT: in vivo reproducibility. Eur Radiol

2004; 14:86–92

14. Kostis WJ, Reeves AP, Yankelevitz DF, Henschke

CI. Three-dimensional segmentation and growth-

rate estimation of small pulmonary nodules in

helical CT images. IEEE Trans Med Imaging

2003; 22:1259–1274

15. Ko JP, Rusinek H, Jacobs EL, et al. Small pulmo-

nary nodules: volume measurement at chest CT—

phantom study. Radiology 2003; 228:864–870

16. Yankelevitz DF, Reeves AP, Kostis WJ, Zhao B,

Henschke CI. Small pulmonary nodules: volu-

metrically determined growth rates based on CT

evaluation. Radiology 2000; 217:251–256

17. Awai K, Murao K, Ozawa A, et al. Pulmonary

nodules: estimation of malignancy at thin-section

helical CT—effect of computer-aided diagnosis

on performance of radiologists. Radiology 2006;

239:276–284

18. Dougherty E. Digital image processing method.

New York, NY: Dekker, 1994:77–85

19. Bland JM, Altman DG. Statistical methods for as-

sessing agreement between two methods of clini-

cal measurement. Lancet 1986; 1:307–310

20. de Hoop B, Gietema H, van Ginneken B, Zanen P,

Groenewegen G, Prokop M. A comparison of six

software packages for evaluation of solid lung

nodules using semi-automated volumetry: what

is the minimum increase in size to detect growth

in repeated CT examinations. Eur Radiol 2009;

19:800–808

21. Goodman LR, Gulsun M, Washington L, Nagy PG,

Piacsek KL. Inherent variability of CT lung nodule

measurements in vivo using semiautomated volu-

metric measurements. AJR 2006; 186:989–994

22. Winer-Muram HT, Jennings SG, Meyer CA, et al.

Effect of varying CT section width on volumetric

measurement of lung tumors and application of

compensatory equations. Radiology 2003; 229:

184–194

23. Kostis WJ, Yankelevitz DF, Reeves AP, Fluture

SC, Henschke CI. Small pulmonary nodules: re-

producibility of three-dimensional volumetric

measurement and estimation of time to follow-up

CT. Radiology 2004; 231:446–452

24. Marten K, Auer F, Schmidt S, Kohl G, Rummeny

EJ, Engelke C. Inadequacy of manual measure-

ments compared to automated CT volumetry in

assessment of treatment response of pulmonary

metastases using RECIST criteria. Eur Radiol

2006; 16:781–790

25. Schwartz M. A biomathematical approach to clin-

ical tumor growth. Cancer 1961; 14:1272–1294

26. Noguchi M, Morikawa A, Kawasaki M, et al.

Small adenocarcinoma of the lung: histologic

characteristics and prognosis. Cancer 1995;

75:2844–2852

27. Ikeda K, Awai K, Mori T, Kawanaka K, Yama-

shita Y, Nomori H. Differential diagnosis of

ground-glass opacity nodules: CT number analy-

sis by three-dimensional computerized quantifi-

cation. Chest 2007; 132:984–990

28. Das M, Muhlenbruch G, Katoh M, et al. Auto-

mated volumetry of solid pulmonary nodules in a

phantom: accuracy across different CT scanner

technologies. Invest Radiol 2007; 42:297–302

29. Goo JM, Tongdee T, Tongdee R, Yeo K, Hildebolt

CF, Bae KT. Volumetric measurement of synthetic

lung nodules with multi-detector row CT: effect of

various image reconstruction parameters and seg-

mentation thresholds on measurement accuracy.

Radiology 2005; 235:850–856

30. Gietema HA, Schaefer-Prokop CM, Mali WP,

Groenewegen G, Prokop M. Pulmonary nodules:

interscan variability of semiautomated volume

measurements with multisection CT—influence

of inspiration level, nodule size, and segmentation

performance. Radiology 2007; 245:888–894